PhD Course: Mathematical Optimization for Machine Learning

The course is aimed at providing basic Numerical Optimization tools to handle some classes of Machine Learning problems, particularly focusing on supervised classification. Optimality conditions for functions of several variables both in the constrained and the unconstrained case will be briefly recalled, along with some fundamental notions in convex analysis. The problem of separating sets in n-dimensional spaces by appropriate separation surfaces will be put in the form of an optimization problem. The Support Vector Machine (SVM) approach, where separating hyperplanes are adopted for classification purposes, will be discussed, together with possible alternative separation methods based on piecewise affine, ellipsoidal, and spherical surfaces.

Multiple-Instance classification models will be discussed as well, together with the results of some applications.

Seminar room - 5th floor, cube 42C

15/04/2024 (09:30-12:30)
16/04/2024  (09:30-12:30)
18/04/2024  (10:00-12:00)

8 ore 2 cfu